The rise of radarAmong all technologies applied to the

autonomous vehicles, different sensors

are utilized to sense the environment.

Lidar, radar, sonar and optical sensors

act as the ‘digital eyes’ of an autono-

mous vehicle. Some modern vehicles

have already been equipped with a vari-

ety of these sensors. For each different

sensor, corresponding algorithms are

required to provide the information for

an advanced control system that inter-

prets the information and determines

strategies in different scenarios, such as

identifying the appropriate navigation

paths in order to avoid obstacles.

Among the sensors, radar is becoming

a key element for autonomous vehicles

due to its all-weather and day-and-night

capabilities. Radar outperforms optical

sensors in low-vision conditions and se-

vere weather. Also, unlike a Lidar sensor,

radar can provide precise velocity meas-

urements. Recent advancements in the

semiconductor industry have made the

low-cost mass production of single-chip

automotive radars possible. Millime-

ter-wave radar is the most widely used

sensor for automotive radar systems,

which has been well-developed in the

past decades and has been widely used

in current Advanced Driver Assistance

Systems (ADAS). Û

Dr. Faruk Uysal

Autonomous systems have been subject to massive developments, especially with the increasing interest in auton-omous vehicles. A fully autonomous vehicle (such as a self-driving car) must be capable of sensing its environment and moving without any human intervention. Some of the current vehicles are already being deployed with auton-omous functionalities, such as autonomous parking, collision avoidance and even auto-pilot systems that support drivers in their task to increase traffic safety. Nevertheless, until a vehicle can drive truly independently, it is not genuinely an autonomous vehicle.

The Digital Eyes of Autonomous Vehicles

Autonomous Systemsand RADAR

Figure 1. A typical fully integrated RFCMOS Radar Transceiver (right) for automotive ra-dar application, shown in a test setup in an electromagnetic anechoic chamber.

January 2019 21

MA ELL 22.2

The Frequency-Modulated Continuous

Wave (FMCW) radar system is the most

common automotive radar system to

detect the range and velocity of targets

through stretch processing. Moreo-

ver, recent automotive radar systems

are taking advantage of multiple-in-

put-multiple-output (MIMO) antenna

arrays to provide the azimuth informa-

tion of targets. Depending on the MIMO

antenna configuration, it is also possible

to exploit the elevation information of

the targets.

Challenges in Automotive Radar Current mm-Wave automotive radar

sensors (such as shown in Figure 1)

share a spectrum space from 76 to 77

GHz (up to 81 GHz in some geographi-

cal regions) [1]. Soon, the co-existence of

multiple radars in congested traffic will

be an issue with the increasing number

of radar-equipped vehicles on the roads.

Since a lot of equal or similar waveforms

and transmission strategies are pres-

ently used in automotive radar applica-

tions, interference will occur between

multiple radar units.

Under the influence of interference,

objects with low radar cross sections

(RCS), such as pedestrians or cyclists,

will not be detected or will be com-

pletely lost during tracking. Therefore,

interference will lead to dangerous sit-

uations and will become a bottleneck

for driving assistance and autonomous

vehicles. Especially in fully autonomous

vehicles, the dependability on the sen-

sors is extremely high and there is abso-

lutely no tolerance for sensing failures

since any human intervention will no

longer be present.

Figure 2 shows a typical automotive

radar output for range versus velocity

domain, which is used to estimate the

range and velocity of the object in the

scene of interest. A received signal that

is under the influence of strong interfer-

ence, is illustrated in Figure 2a. As seen

from the figure, it is not possible to de-

tect a target, since it is masked by the

interference. Figure 2b and Figure 2c

show the interference and target com-

ponents of the input signal, which are

achieved through the use of a signal

separation algorithm for interference

mitigation as proposed at [2].

Trends and research directions Besides providing range, velocity and

angle information of the targets, radar

systems are often used for the classifi-

cation of targets and/or their activities.

Most of the real world targets are not

rigid bodies. Micro-motions or vibra-

tions induced by different parts of the

targets produce additional Doppler

shifts, which is known as micro-Doppler

effects [3], and can be used to identify

target features. For instance, micro-mo-

tions induced by human body parts

produce a unique micro-Doppler signa-

ture which can be used to identify hu-

man activities. A typical micro-Doppler

signature of a walking person as seen

by an automotive radar is shown after

time-frequency analysis in Figure 3.

Even though radar is one of the key sen-

sors for autonomous systems, there is

still a need for other sensors to achieve

maximum safety and security. This is

6

Fig. 5. Real data processing: a) collected signal, b) collected signal after pre-processing c) signal o�nterest and d) interference afterprocessing.

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0410

6

Fig. 5. Real data processing: a) collected signal, b) collected signal after pre-processing c) signal o�nterest and d) interference afterprocessing.

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0410

6

Fig. 5. Real data processing: a) collected signal, b) collected signal after pre-processing c) signal o�nterest and d) interference afterprocessing.

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0410

Figure 4. Multimodal machine learning for human activity classification [4]..

Figure 2. Typical output of an automotive radar at range vs velocity domain. a) received radar signal, b) interference signal and c) target signature after interference mitigation [2].

January 201922

MA ELL 22.2The Digital Eyes of Autonomous Vehicles

because each sensor has its own advan-

tages and disadvantages. The most ap-

propriate approach to overcome the in-

herent weaknesses of different sensors

is to deploy a combination of different

sensors and fuse their data before the

decision-making process. Most of the

current systems use late-fusion strat-

egies, which combine the final product

from each sensor. However, there is a

great interest in early-fusion strategies

such as multi-modal machine learning,

which aimsto train a network using dif-

ferent sensor data such as radar and op-

tical sensors as illustrated in Figure 4.

Other works in Progress The Microwave Sensing, Signals and

Systems (MS3) Group at the Faculty of

Electrical Engineering Computer Sci-

ence and Mathematics (EEMCS) at Delft

University of Technology is collaborat-

ing with NXP Semiconductors N.V. in

the “Coded-Radar for Interference Sup-

pression in Super-Dense Environments”

(CRUISE) project to tackle issues relat-

ed to spectrum crowding. CRUISE will

fully support the future of autonomous

driving by exploiting spread-spectrum

techniques to ensure proper radar sig-

nal detection and object classification in

a highly-occupied frequency spectrum,

as well as accurate ranging, velocity and

azimuth measurements under all cir-

cumstances.

Moreover, the MS3 group is working

towards the ”Integrated Cooperative

Automated Vehicle1” (i-CAVE) to ad-

dress the current challenges regarding

throughput and safety with an integrat-

ed approach to automated and coop-

erative driving. In the i-CAVE project,

radar-based communication will be

realized to achieve a more robust and

synergetic approach for joint sensing

and communication during high-speed

automated and cooperative driving.

To address interaction capabilities be-

tween vehicles and the environment,

our research focuses on radar process-

ing methods with signals that allow for

communication functionality.

Ê

“Hopefully in the near future, radar will not only be the digital eyes, but also the digital voice of

autonomous systems.”

[1] J. Hasch, E. Topak, R. Schnabel, T. Zwick, R. Weigel, and C. Wald- schmidt, “Millimeter-wave technology for automotive radar sensors in the 77 GHz frequency band,” IEEE Transactions on Microwave Theory and Techniques, vol. 60, pp. 845–860, March 2012.

[2] F. Uysal and S. Sanka, “Mitigation of automotive radar interference,” in 2018 IEEE Radar Conference (RadarConf18), pp. 0405–0410, April 2018.

[3] V. C. Chen, The Micro-Doppler Effect in Radar. Artech House, 2010. [4] R. de Jong, “Multimodal deep learning for the classification of human activity,” Master’s thesis, Delft University of Tech-

nology, 2019.

0.5 1 1.5 2 2.5 3

time (s)

0

0.5

1

1.5

2

2.5

3

3.5

4ve

loci

ty m

/s

Figure 3. A walking human’s micro Doppler signature as seen by an automotive radar. Swinging arms and legs create a unique signature since they moved with a different velocity..

January 2019 23

MA ELL 22.2The Digital Eyes of Autonomous Vehicles